Evidence before folklore
Music mastering guides for decisions you can hear.
Plain-English explanations of loudness, peaks, formats, stereo translation and generated-audio repair. Each guide states its method, links its sources and separates platform guidance from universal rules.
Start here
Mastering for AI-generated music: improve the sound, not the story
Generated audio can arrive with strong musical ideas and difficult technical baggage. This workflow focuses on audible quality while keeping authorship and detector claims honest.
Read the workflowCORE PRINCIPLE
Improve the audible result without inventing a claim about how the music was made.
The library
Prepare, measure, repair and deliver.
Streaming delivery
Spotify loudness, LUFS and true peak without the myths
Spotify publishes a default playback reference of -14 LUFS, but that number describes normalization behaviour, not a universal rule that every master must hit.
Read guide 02Pre-master checklist
How to prepare a mix for mastering
A good pre-master is not artificially quiet or stripped of character. It is the best version of the mix, exported cleanly with enough information to make the next decision.
Read guide 03Audio formats
WAV vs FLAC vs MP3 for mastering
WAV and FLAC can both preserve audio without perceptual loss. MP3 is designed to reduce data, making it useful for listening copies but a poor first choice for a mastering source.
Read guide 04Peak control
Clipping, sample peaks and true peak
A meter can reveal overload risk, but the waveform and the listening context determine whether the problem is an accidental clip, intentional saturation or an inter-sample peak.
Read guide 05Stereo translation
Stereo phase and mono compatibility
Width is valuable when it survives real playback. Correlation and mid/side meters can guide inspection, but mono listening reveals what the listener will actually hear.
Read guide 06Repair before polish
How to repair common AI-music audio artifacts
Generated tracks can contain familiar audio problems and model-specific textures. Identify what is actually audible, then use the least destructive repair that helps.
Read guideEDITORIAL METHOD
Useful, sourced and open about limits.
We start with official standards, current platform documentation and primary research. We distinguish measurements from listening judgments, identify third-party claims, and date every review. We do not promise platform acceptance, pressing approval, “human-made” status or a detector outcome.
ITU-R BS.1770 AES Loudness Project AI-music detection research